PearlLeeStudio/TheArtist-MusicTransformer-ft-pop80

TheArtist Music Transformer — F1 (Pop 10K Mix, pop-leaning)

F1 slot of the pop→jazz mix-ratio sweep: the Phase-0 pop baseline fine-tuned on jazz with a 10,000-sequence pop rehearsal buffer, the pop-leaning endpoint. One of six checkpoints released alongside the paper Empirical Study of Pop and Jazz Mix Ratios for Genre-Adaptive Chord Generation (Lee, 2026), and the base model the 11 per-genre lora-* adapters were trained on and pair with. Recommended for chord-composition workflows targeting pop, rock, CCM, K-pop, J-pop, and modern country with optional jazz coloration; F4 (ft-pop29) is the symmetric jazz-leaning endpoint and F3 (ft-pop50) the balanced middle.

Paper · Code · Demo · All models

Weights note. A full SHA-256 comparison over the 25,841,152-element serialized state dict (the model has 25,661,440 unique parameters; the difference is the tied input/output embedding stored as two tensors) shows the released best.pt is weight-identical to the Phase-0 pop baseline. Best-checkpoint selection used minimum loss on a pop-dominated validation mix, which rose monotonically during jazz fine-tuning, so the pre-fine-tune initialization was retained — and with the run's two-epoch warmup that retained epoch is bit-identical to Phase-0 (a seed/data-dependent outcome, the worst case observed for this seed-42 run, not a general property of the selection rule — matched-data multi-seed retrains selected already-adapted epochs for all three seeds; details on the v2 card). The pop/jazz table below therefore reports the run's best-epoch training-log metrics, while the released weights behave exactly like Phase-0 (jazz top-1 72.86 on the same 6-source test). A selection-corrected retrain is released as ft-pop80-v2 (hash-distinct, selected on a jazz-only validation subset). This repository stays as-is: the 11 lora-* adapters were trained on and pair with this exact base, so their reported gains are gains over a pure-pop harmonic prior. The earlier claim that this base "retains a richer harmonic vocabulary" than the pop baseline is withdrawn.

Model details

Field	Value
Architecture	Music Transformer with relative positional attention
Parameters	25,661,440
Vocabulary size	351 tokens
Max sequence length	256
d_model / heads / FFN / layers	512 / 8 / 2048 / 8
Fine-tune resumed from	Phase-0 pop baseline
Best epoch	6 (training-log; the released best.pt is the epoch-3 minimum-mixed-val selection, see the weights note)

Usage

Requires torch, huggingface_hub. The repo bundles model.py and tokenizer.py, so nothing needs to be cloned from GitHub.

import sys
import torch
from huggingface_hub import snapshot_download

# Download the full repo (model.py, tokenizer.py, best.pt, config.json).
ckpt_dir = snapshot_download(repo_id="PearlLeeStudio/TheArtist-MusicTransformer-ft-pop80")
sys.path.insert(0, ckpt_dir)  # so the next two imports resolve

from model import MusicTransformer
from tokenizer import ChordTokenizer

tokenizer = ChordTokenizer()
ckpt = torch.load(f"{ckpt_dir}/best.pt", map_location="cpu", weights_only=False)
model = MusicTransformer(
    vocab_size=tokenizer.vocab_size,
    d_model=512, n_heads=8, d_ff=2048, n_layers=8,
    max_seq_len=256, dropout=0.0, pad_id=tokenizer.pad_id,
)
model.load_state_dict(ckpt["model_state_dict"])
model.eval()

# Prompt = ii-V-I in C major; ask for a pop-flavoured continuation.
song = {
    "key": "Cmaj", "time_signature": "4/4", "genre": "pop",
    "bars": [["Dm7", "G7"], ["Cmaj7"]],
}
prompt_ids = tokenizer.encode_sequence(song)[:-1]
ids = torch.tensor([prompt_ids])
with torch.no_grad():
    for _ in range(32):
        logits = model(ids)
        next_id = torch.multinomial(
            torch.softmax(logits[:, -1, :] / 0.8, dim=-1), 1,
        )
        ids = torch.cat([ids, next_id], dim=-1)
        if next_id.item() == tokenizer.eos_id:
            break
print(tokenizer.decode(ids[0].tolist()))

For per-genre adaptation beyond pop and jazz, see the 11 LoRA adapter repos at PearlLeeStudio — they chain on top of this base.

Evaluation

Held-out per-genre test sets — the figures below are the fine-tuning run's best-epoch (epoch 6) training-log metrics, with the jazz column measured on the 6-source jazz test (167 sequences); the released weights do not embody them (see the weights note above).

Metric	Pop test	Jazz test (6-src)
Top-1 accuracy	84.60%	81.03%
Top-5 accuracy	96.96%	92.41%
Perplexity	1.78	2.31
Δ vs. Phase-0 baseline	+0.39	+8.17

Out-of-distribution per-genre baseline

F1 alone (no LoRA) on the 11 per-genre val splits the LoRA adapters target (10 Chordonomicon genre subsets plus the Bach-chorale classical split). The eight genres beyond the base vocabulary are encoded at [GENRE:none]-initialised embedding rows (effectively unconditioned); rock, blues, and bossa use their existing base-vocab [GENRE:X] tokens. This is the no-LoRA reference reported on every lora-<genre> adapter card — per the weights note, these are measurements of the released weights and are cleanly interpretable as a pure-pop prior.

Genre	Val seq.	F1 top-1 (%)	F1 top-5 (%)	F1 val loss
hip-hop	1,402	86.51	96.27	0.6240
electronic	1,519	84.50	95.93	0.6835
rock	4,891	82.79	96.75	0.5865
folk	6,075	82.66	95.80	0.7406
funk	283	82.54	94.38	0.7878
country	6,173	82.45	96.22	0.7402
rnb/soul	955	82.09	94.12	0.8119
blues	994	81.70	94.80	0.8137
gospel	374	79.34	94.73	0.8813
bossa	1,431	78.33	93.64	0.9635
classical	37	43.54	72.82	2.8653

Per-genre real-song eval

On this eval set the released F1 base peaks on hip_hop (90.66%) and struggles most on classical (49.55%). The 11 per-genre LoRA adapters (sister lora-* repos) are the recommended path beyond pop and jazz — for the eight genres without a [GENRE:X] token in the 351-token vocabulary (all but rock, blues, and bossa) the base decodes without genre conditioning here.

Genre	n_songs	Top-1 (%)	Top-5 (%)	val_loss
pop	10	86.68	96.01	0.5734
rock	10	86.69	97.48	0.4578
jazz	10	64.96	81.16	1.8958
blues	10	81.52	93.91	0.8410
bossa	10	81.43	95.47	0.7825
classical	10	49.55	81.17	2.2389
country	10	85.90	98.44	0.5152
electronic	10	87.39	98.45	0.5072
folk	10	85.04	98.92	0.5244
funk	10	83.85	96.03	0.6811
gospel	10	79.79	96.85	0.7367
hip_hop	10	90.66	98.59	0.3957
rnb_soul	10	85.10	97.07	0.5877

130 songs (10 per genre × 13 genres, seed 42) drawn from held-out val/test partitions — pop from McGill Billboard (CC0), jazz from public standards corpora, classical from Bach chorales, the other ten genres from the matching Chordonomicon subsets (CC BY-NC 4.0; titles are Spotify track IDs by upstream policy). Full composition:

Genre	n	Source(s)	Bar range	Avg duration · named
pop	10	billboard	58–116	189s · 10/10 named
rock	10	chordonomicon_rock	52–87	127s · 0/10 named
jazz	10	choco:jazz-corpus, choco:real-book, jazzstandards, jht	16–89	72s · 10/10 named
blues	10	chordonomicon_blues	24–46	93s · 0/10 named
bossa	10	chordonomicon_bossa	24–78	88s · 0/10 named
classical	10	chordonomicon_classical	11–40	60s · 10/10 named
country	10	chordonomicon_country	30–81	110s · 0/10 named
electronic	10	chordonomicon_electronic	25–84	89s · 0/10 named
folk	10	chordonomicon_folk	33–82	114s · 0/10 named
funk	10	chordonomicon_funk	30–60	92s · 0/10 named
gospel	10	chordonomicon_gospel	24–85	98s · 0/10 named
hip_hop	10	chordonomicon_hip_hop	24–81	136s · 0/10 named
rnb_soul	10	chordonomicon_rnb_soul	34–82	128s · 0/10 named

Source license summary: McGill Billboard (CC0, named pop songs), Jazz Harmony Treebank / JazzStandards / WJazzD (Public / community-redistributed, named jazz standards), Bach chorales via music21 (public domain, named pieces), Chordonomicon per-genre subsets (CC BY-NC 4.0; titles are Spotify track IDs by upstream dataset policy — progressions are real songs).

Training data

All 1,513 jazz training sequences (Jazz Harmony Treebank, JazzStandards, Weimar Jazz Database, JAAH) plus 10,000 pop rehearsal sequences sub-sampled with seed 42 from the Phase-0 pop training split — pop:jazz ≈ 6.6:1 in the mix. Fine-tune hyperparameters: peak learning rate 2 × 10⁻⁵, two-epoch warmup, ten epochs maximum with patience 5.

License

CC BY-NC 4.0 (weights; matching Chordonomicon, the dominant training corpus). Research, paper replication, portfolio, and demo use are permitted; commercial use is not.

Citation

@misc{lee2026chordmix,
  title         = {Empirical Study of Pop and Jazz Mix Ratios for Genre-Adaptive Chord Generation},
  author        = {Lee, Jinju},
  year          = {2026},
  eprint        = {2605.04998},
  archivePrefix = {arXiv}
}

@misc{lee2026chordtimeseries,
  title         = {How Far Can Chord-Symbol Time-Series Adaptation Carry Genre Identity?},
  author        = {Lee, Jinju},
  year          = {2026},
  eprint        = {2606.07334},
  archivePrefix = {arXiv}
}